It Doesn't Look Like Polls Are Underselling Biden
Election Update, June 2024
Not much has changed in polling since we last checked in in March: Trump is still ahead nationally. Trump leads not only in the swing states that will decide the electoral college, but in the national popular vote (albeit by a much narrower margin).
Personally, I expected Trump’s felony convictions to bring him down by a couple of points. (I also thought he would recover a few weeks later, but I thought we would see an impact.) There was some early evidence that the conviction would indeed hurt him, but, nearly three weeks later, the conviction has left no obvious impact on the polls at all. You can kinda see something if you squint? In the seven days leading up to the conviction, Trump led the 538 national average by an average of 1.2%. In the seven days leading up to this post, Trump led by 1.0%. So he has lost ground since the conviction… but that’s probably just normal tiny fluctuations in the statistical average. (For example, in the random seven-day period May 8 - May 15, Trump led by an average of 0.8%. Movement of less than half a point is usually just statistical noise.)
Either way, whether the conviction’s impact was marginal or actually non-existent, it seems likely that, after two and a half weeks, whatever impact it’s going to have on the race is spent. If low-information voters were going to react, they likely would have by now. They haven’t. Thus, nothing has really changed since March.
Trump is leading because (as we discussed in March) he has made huge inroads with non-whites, who have voted monolithically for Democrats for a long time.1 However, Biden is holding his 2016-2022 gains among educated suburban whites who are repelled by Trump’s personality, his personal peccadillos, and his penchant for trying to overthrow the Constitution.2 (Also, suburban whites are big mad about their convenience abortions.)
This creates a lot of uncertainty about the election outcome. Educated suburban whites, who are increasingly behind Biden, are extremely reliable voters. Non-whites overall, who are warming up to Trump, are some of the least reliable voters. This raises an interesting strategy question: is it better to hold a narrow lead that depends on voters who might not even show up, or is it better to have voters you can really count on, but a smaller number of them?
The first one. Unquestionably. Zeynep’s Law applies. It’s an election. Bigger number good, smaller number bad.
Nevertheless, Trump’s support is soft enough to create a lot of uncertainty, which is so far confounding the election modelers. The Economist thinks we can pretty much trust what people are saying in the polls, even though people are manifestly untrustworthy in polls all the time. Therefore, they give Trump a meaningful (though hardly conclusive) lead.3 538 thinks we should lean more on economic fundamentals, special elections, and so forth, even though polls are manifestly more predictive than those other measures at this stage in a race. Therefore, 538 rates the race as effectively tied.4 When there are only wrong answers, which wrong answer should we trust more?
Neither! 538 (née FiveThirtyEight5), used to be the gold standard in election forecasting, but, last year, they lost their founding editor / greatest living election forecaster, Nate Silver, and all his forecasting models. They had to hire G. Elliot Morris, an outspokenly partisan Democrat who swiftly kicked the Rasmussen Poll out of the 538 average for being too right-leaning.6 I do not have confidence in his abilities. Morris was hired away from The Economist, forcing The Economist to hire a statistician named Andrew Gelman to make their model. Gelman has published academic work in this area but who has never run a national model before.7 As a result, neither model has a good track record and so neither can be considered a reliable predictor. The old FiveThirtyEight models, which did have a proven track record, are all still owned by Nate Silver, who is launching a 2024 forecast later this month on his new Substack, Silver Bulletin:
I’m on tenterhooks waiting for the new forecast. My educated guess is that Nate will land between 538 and The Economist, but a little closer to 538. (Trump 60% or so.) But I am going by my gut when I say that, and I trust Nate’s numbers over my gut every time. I have been rewarded many times for following the “trust Nate Silver” strategy in every cycle since 2008.8
Until then, though, I’m a double-hater who won’t vote for either the disqualified insurrectionist or the proud child-killer, so I’ve had very little to do this election cycle except read polls and watch partisans try to rationalize bad results.9 Making others as miserable and misanthropic as I am is the only bitter pleasure left in the wreckage of my political life.
Most of the rationalizations are obvious. For instance, anyone who tells you “polls aren’t predictive” this early or “polls are meaningless now” because response rates have gotten so low has no clue what he is talking about. Even notoriously fiddly primary polls are fairly predictive even a year in advance, and 2022 showed that polls are still accurate. You can blast that hopium out the window with a bog-standard blow dryer.
There are other rationalizations, though, that are a little more interesting. My favorite is the theory that Joe Biden is actually ahead right now, but the polls can’t see it. I sincerely think this is plausible on its face. Here is how the theory runs:
For a long time, pollsters were happy campers. After a bad miss in 1998, where they missed the top-of-the-ticket Senate races by (on average) an enormous 4.5 points, pollsters cleaned up their acts:
For the next twenty years, pollsters collectively kept their overall partisan error at or below 3 points. After three remarkable election cycles in a row (2006-2010) where the polls were essentially bang-on, pollsters finally made a modest mistake in 2012, when they expected Romney to lose by 1.4 points, a narrow escape for Obama. Romney instead lost by a convincing 3.9 points. However, pollsters swiftly corrected—in fact over-corrected! In the subsequent 2014 cycle, they overestimated Democrats by the exact same margin.
Another modest but harrowing mistake came in 2016. Pollsters’ overall error hit the 3-point ceiling, and they missed the presidential result by 3.3 points. This time, everyone noticed, because, unlike the similar miss in 2012, this one changed the outcome: instead of Trump losing a close race, Trump won a close race. But the miss was still only moderate, and, again, pollsters corrected: they were almost perfect in 2018’s midterms.
Then, finally, 2020 came, and it was a reckoning. Biden was expected to romp to a win, ahead by 8.6 points, which would crush Trump and Trumpism forever. Instead, he won by 4.5 points, which was just barely enough to overcome Trump’s (unusually large) advantage in the electoral college. Biden won, but by the skin of his teeth. Downballot polling results were significantly worse. It was the worst polling in a generation, a real whiff.
Pollsters had some good reasons for struggling in 2020: covid, crazy balloting rules changing on the fly, race riots, unusually awful candidates in marquee races. Still, pollsters would be crazy to not make corrections after a miss that big, especially since it similar to their miss in 2016, which they thought they had already fixed!
According to the pro-Biden theory of polls, that’s exactly what pollsters have done. They’ve corrected for all the Trump voters they missed in 2020… but they have overcorrected, and are showing Trump doing better than he really is. Biden is actually ahead right now (according to this theory), but polls can’t see it because they’re focusing too much on Trump voters. People who hold this theory can even point to a precedent: the 2012 polling miss was followed by an equal-but-opposite polling miss in 2014.
There’s precious little direct evidence for this rationalization, besides vibes. The main point cited in its favor is the fact that Democrats keep beating the polls in special elections this year (which is true)—but that seems to be explained by the fact that Democrats have acquired the reliable educated suburban voters who used to deliver special elections to Republicans. Still, unlike a lot of rationalizations, this one is at least plausible! Pollsters do adjust their methods between cycles and they can overadjust. If Biden does indeed win in 2024 (which even The Economist says has a 25% chance of happening), we may very well look back in retrospect and realize that this is exactly what happened.
However, there are some good reasons to doubt this theory, one of which has come more sharply into view in recent weeks.
The first reason is the 2022 midterm election. The mainstream narrative about the 2022 midterm election is that it was a big miss for polls, where pollsters favored Republicans by big margins and created a “red wave” mirage. This narrative supports the idea that pollsters overcorrected after 2020. However, this narrative has been obviously false from the very beginning. 2022 was a big miss for pundits, who insisted on a red wave even though the polls showed a red splash. Pundits blamed polls for their own stupidity, and somehow got away with it. In reality, 2022 was a fantastic election for polls, which were very nearly dead-on. (I’ve written quite a bit about this.) So, actually, 2022 gives us some evidence that whatever adjustments the pollsters made after 2020 were just right, not an overcorrection. Indeed, given how much the GOP is now relying on unlikely voters (who turn out way more in presidential years than midterm years), one could make the case that, if anything, 2022 showed that pollsters have not made enough adjustments to account for the much broader presidential electorate expected in 2024.
A more damning problem for the “overcorrection” hypothesis is the actual polls we’ve been getting in the 2024 cycle.
There are certain pollsters who are just really, really good. They have a track record of being canaries in the coal mine, breaking out of the herd, and showing the world what’s actually going on.
In October 2022, the NYT/Siena College poll posted results that their own columnists admitted seemed completely unbelievable: huge Democratic inroads in districts that should be tilting red! Weaker-than-expected GOP leads all over the place! Everyone complimented them for their bravery in publishing results that might very well turn out to be outliers. Of course, they turned out to be right: the GOP did have weaker leads than expected, and had a disappointing midterm. NYT/Siena does this routinely.
In particular, NYT/Siena probably did not need to make as many significant adjustments after 2020, because NYT/Siena did relatively well in 2020. If they didn’t make as many major adjustments, they probably didn’t overcorrect, or at least didn’t overcorrect as much. If the “overcorrection” hypothesis is correct, then, Biden should probably already be consistently ahead in the Siena Poll today, even though the consensus of less-accurate pollsters might (inaccurately) show Trump with a slight lead.
Instead, NYT’s latest poll (restricted to battleground states) showed Trump up by 6 points, which is worse than the consensus of Trump being up by 2-3 points in the battlegrounds.
The famous Selzer Poll, run by J. Ann Selzer for the Des Moines Register, (usually) only polls one state: Iowa. But ohhhhh does it ever poll Iowa well. Repeatedly throughout the years, Selzer has posted results that defied conventional wisdom—and the consensus of other polls: in 2014, she said Joni Ernst was up by 7 when other polls said the race was tied. Ernst won by 8.5. In 2020, the polling consensus said Iowa was close—Trump was winning, but only by 1-2 points—but Selzer said Trump was up 7. Trump won by 8.10 In 2012, the final polls excluding Selzer (which still included some very good pollsters) showed Obama ahead by 2. Selzer showed Obama winning by 5. He won by 5.8.
Selzer just released a poll saying that Trump is ahead in Iowa today by 18 points.
You may think, “Okay, so what? Everybody knows Trump is going to win Iowa in 2024, so who cares how big Trump’s lead is? It’s 6 electoral votes whether he wins by a million votes or one vote.”
The reason this is scary news for Biden is because Iowa bears a close demographic resemblance to at least the rural areas of other states… including battleground states like Minnesota and (especially) Wisconsin, a nearly-vital state for Biden. On the night in 2020 when Trump won Iowa by 8, he essentially tied in Wisconsin. If Trump wins Iowa by 18, there is very little doubt that he will win Wisconsin, and perhaps even Minnesota. Indeed, if correct, this poll shows a 10-point shift against Biden since 2020. A 10-point shift against a presidential candidate in any state tends to be replicated by a significant shift against that candidate in every state, demographically similar or not. Some people dismiss Iowa polling because so many Iowans are White… but you know what every single American state has in common? White people! Lots of ‘em!
So, as Silver succinctly put it yesterday, “It’s not a great sign for Biden that some of his worst numbers come from the best pollsters.” There’s a reason NYT/Siena and Selzer are the two highest-rated pollsters in Silver’s legendary pollster ratings. If these two “canaries” are signaling anything, it’s that polls might not be showing enough of a Trump lead right now.
I wouldn’t bet money on that. A+ pollsters are wonderful, but, in the long run, you’re better off with a model that takes account of all data from all pollsters, instead of just taking the A+ pollsters’ word as gospel. Trump is probably not ahead in Iowa by 18 points. However, I think these A+ results, combined with polls’ high accuracy in 2022, do give us sufficient reason to dismiss the “overcorrection hypothesis,” at least given current evidence.
Of course, there are plenty of other reasons why pollsters might be getting 2024 wrong, in either direction! Trump is only a normal polling error ahead of Biden, and, even if there weren’t still a ton of time left before the election, we know how a normal polling error worked out in 2016. I do not think this means Trump has it in the bag, especially not with the debates still to come.11 In 2022, polling margins fluctuated by over 4.2 points between June and October, and Trump’s electoral college lead would be erased and reversed by a 4-point swing.
If you want a sports analogy, we can compare this election to a baseball game: I think Trump is beating the “home team” (the incumbent) by 1 run, with no outs and a runner on first… but it’s only the top of the second inning. He’s in a good position, but there’s plenty of game left to play.
However, I do think we have good reason to believe, at this point, that pollsters are not missing a secret Biden lead due to a post-2020 overcorrection. Trump is ahead.
At least for today.
At least if his voters show up to the polls.12
UPDATE 20 June 2024
It’s possible I spoke too soon about the conviction not having an impact! On June 16, just before I started writing this post, Trump had a 1.1-point lead in the 538 average. Now, just four days later, Biden has had a string of good polls and now leads the average by 0.1 points! This is the thinnest of technical leads and doesn’t change Trump’s lead in the swing states, but it’s a big enough swing in a short enough time to make me think something real might be going on, not random polling fluctuations. It’s also notable because it’s the first time Biden has led in polling in at least nine months. I’m keeping tabs.
Hispanics have been a Democratic monolith since Obama brought all non-white demographics together in 2008 and 2012. George W. Bush believed they were the future of the Republican Party and made huge strides with them in 2004, but Obama turned that promising start into a high-water mark.
Blacks have been a Democratic monolith since the New Deal. Most people think that the Democrats won the Black vote by abandoning Southern Democrats and embracing the Civil Rights Movement, for very understandable reasons… but that’s not actually true. The Party of Lincoln lost the Black vote to FDR almost 20 years earlier and never recovered. Make of that what you will. I’d like to write about it more on some other occasion.
Asians were such a small part of the electorate that they weren’t even measured in national exit polling until 1992 — and were actually right-leaning in presidential races until 2000, when they started to become more solidly Democratic.
Arabs have always been, and remain, a pretty small category of American voters, but Muslims have babies and nobody* else does, so they are getting bigger. (*except Mormons, the Amish, and the slender remnant of orthodox Roman Catholics) If I recall, Arabs were trending toward Republicans in the late 1990s over social issues. To say that Muslims aren’t big on abortion or gay pride is an understatement. However, 9/11 and everything that came after, from the Iraq War to Jack Bauer, decisively pushed Arabs away from Republicans for a generation. Arab voters are furious right now with Democrats over their tepid support for Israel, but it’s not clear that this is driving their movement toward Republicans, since Republicans support Israel much more strongly.
Trump supporters sometimes sum these up as “being mad about mean tweets,” which is kind of like calling the international despairing nihilism that gripped the world in the 1920s, after whole generations of young men were slaughtered wholesale at the Somme and elsewhere in World War I, “being mad about a mean telegram.”
For those reading this from the future: as I write this at 3:52 PM Central on June 17 from the New Brighton indoor playground where I’ve taken my children on my day off, The Economist currently gives Trump a 72% chance of winning. They seem to be counting electoral college ties as Trump wins, which is a defensible choice.
For those reading from the future: 538 gives Trump a 50.2% chance of winning, Biden a 49.4% chance of winning, and a 0.4% chance of a tie. (Note: Trump very very likely wins an electoral college tie, p>95%.)
Hopefully this won’t come up too often, but I will generally use FiveThirtyEight to refer to the Nate Silver golden age of the website, and 538 to refer to the G. Elliot Morris era, since they conveniently changed the styling right around when Morris came on board.
That’s obviously not what Morris claimed, but there’s good evidence that he’s lying—although, as usual, it’s more likely that he’s lying to himself than that he’s consciously lying to us. Many such cases.
Not that I know of, anyway. I have few opinions on Gelman, who keeps a relatively calm public profile and used to write generally reasonable, often correct stuff for The Monkey Cage at the Washington Post. However, I have been Nate-Silver-or-Bust on presidential election modeling since 2016, Nate Silver’s finest hour.
I guess I’ve also got theological hot takes on both candidates, but it’s hard to find an audience for this. Both candidates help concretely illustrate points Ed Feser made at a fairly abstract level in his important blog post, “How to Go to Hell.” Both exemplify the conditions of mortal sin—grave evil, full knowledge, and full consent of the will—in a really public way that helps illustrate why, for example, your average teenage wanker has probably not damned himself, whereas it is more likely that Trump and Biden both have. Yet, even while both sin tremendously, they sin in very different ways, with very different attitudes, which helps us to see some of the different paths to Hell. (“All saints are alike; each mortal sinner is unhappy in his own way,” perhaps.)
This has given me a lot to think about for the past few years, but, like I said, who’s the audience? “Partisans rationalizing poll results” has given me way more chances for nerd-sniping on Reddit.
Lest you think that these polls—or any polls—are infallible, it’s worth mentioning here that the NYT/Siena poll, the one I just sang the praises of above, said Iowa was +3 Biden the same week that Selzer correctly reported that Trump was leading by a mile. NYT/Siena did better than the average pollster in 2020, but it was still a shellacking for pollsters and NYT/Siena did not escape unscathed.
Team Trump has stupidly set such low expectations for Biden that, as long as Biden can stand and talk in complete sentences for 90 minutes, I think the content won’t matter; he will impress the nation.
I’ve said it before, I’ll say it again: I can’t wait to watch the partisan valence on ballot access reverse, until Republicans are demanding same-day voter registration and Democrats ban early voting.
Just coming back a month later to lament on how rosey footnote #11 ended up being…
Your TDS is disgusting. You give off the appearance that you would sacrifice the unborn if it stopped Trump from being president.